All of Nicholas Goldowsky-Dill's Comments + Replies

Cool work! Excited to dig into it more.

One quick question -- the results in this post on probe generalization are very different than the results in the paper. Which should I trust?

FYI, See this paper for a more detailed explanation of the model in question and how it's trained.

In the sandbagging task, what is the relationship between task performance and CoT eval recognition? In the example you provide, it considers the sandbagging but doesn't perform it. Is that reflective of most of the trials?

No clear difference. I did a quick check and Sonnet's accuracy at answering the questions is actually lower in transcripts where it's evaluation aware. But this effect seems to be driven by a single sample where the model runs into message limits and thus accidentally scores 0, and I wouldn't expect it to replicate. Looking at the ... (read more)

Yes I'm excited about "probing the answer to followup questions" as a general technique. Our results were promising but there's a lot of iteration one could do to make the technique work better!

Separately, did you run any experiments including the probe on the CoT tokens? I would suspect there to be a pretty reliable signal here, and ideally this could be used to help confirm honest CoT. 

On our website you can look at the probe scores on the CoT. I don't have up to date numbers, but I expect it's mildly helpful for improving classification at le... (read more)

One question I have about the Mechanistic Interpretability safety case sketch from Roger Grosse: How do you actually find features of high concern? 

Features of High Concern are things like "intent to carry out a harmful plan". By definition, they very rarely activate on benign deployment outputs. This must be the case as "if the features [of high concern] activate more than a handful of times in deployment (internal or external), we discontinue the model and roll back to a weaker model." [M1.1]

They propose to find these features [S1.1] by pe... (read more)

Just checking -- you are aware that the reasoning traces shown in the UI are a summarized version of the full reasoning trace (which is not user accessible)?

See "Hiding the Chains of Thought" here.

See also their system card focusing on safety evals: https://openai.com/index/openai-o1-system-card/

Have you looked at how the dictionaries represent positional information? I worry that the SAEs will learn semi-local codes that intermix positional and semantic information in a way that makes things less interpretable.

To investigate this one can take each feature and could calculate the variance in activations that can explained by the position. If this variance-explained is either ~0% or ~100% for every head I'd be satisfied that positional and semantic information are being well separated into separate features.

In general, I think it makes sense to spe... (read more)

Cool paper! I enjoyed reading it and think it provides some useful information on what adding carefully chosen bias vectors into LLMs can achieve. Some assorted thoughts and observations.

1. I found skimming through some of the ITI completions in the appendix very helpful. I’d recommend doing so to others.
2. GPT-Judge is not very reliable. A reasonable fraction of the responses (10-20%, maybe?) seem to be misclassified.
   1. I think it would be informative to see truthfulness according to human judge, although of course this is labor intensive.
3. A lot of the mileage seem

My summary of the paper:

1. Setup
   1. Dataset is TruthfulQA (Lin, 2021), which contains various tricky questions, many of them meant to lead the model into saying falsehoods.  These often involve common misconceptions / memes / advertising slogans / religious beliefs / etc.  A "truthful" answer is defined as not saying a falsehoood. An "informative" answer is defined as actually answering the question.  This paper measures the frequency of answers that are both truthful and informative. 
   2. “Truth” on this dataset is judged by a finetuned version of